// About

The ML team you'd hire if they were available.

Small senior team in Wrocław. We embed with product teams - Brand24, SurferSEO, Comcast, Chili Piper - to ship LLMs, fine-tunes, computer vision, and the eval infrastructure that catches regressions in production.

2022founded
60+projects shipped
16+open-source models
80Kmodel downloads / month
10+peer-reviewed papers
13+hackathon wins

// What we do

Two things, mostly.

  • Production

    Building and running ML systems in production. Agents, RAG, inference optimization, observability, evals, and the on-call rotation that comes with it. Engagements run 2-12 weeks, embedded with a customer's eng lead.

  • Research

    Hands-on ML research and prototyping. Custom fine-tunes (LoRA, QLoRA, DPO, KTO, GRPO), benchmark and eval methodology, synthetic data pipelines. Some of it ships as open source on HuggingFace; some stays inside customer products.

// Leadership

The people deciding what ships.

Karol Gawron

Karol Gawron

Co-founder · Head of R&D

Leads model and inference work. Background in academic NLP; ships the production version of what the team publishes. Forbes 30 Under 30.

LinkedIn
Michał Pogoda-Rosikon

Michał Pogoda-Rosikon

Co-founder · CEO

Engineering and product. Owns customer-facing scoping, eval design, and the parts of the stack that have to actually run. Forbes 30 Under 30.

LinkedIn
Michał Świędrowski

Michał Świędrowski

Co-founder

Commercial and partnerships. Translates between what customers say they want and what actually moves the metric.

LinkedIn
Norbert Ropiak

Norbert Ropiak

Co-founder

[placeholder bio - replace with actual role and one-line description].

LinkedIn

// Timeline

What got us here.

  • 2022

    Founded out of academic NLP research

    Spun out of the Polish NLP research community. Initial team was published researchers who wanted to ship.

  • 2023

    First production engagements

    Brand24, SurferSEO, Comcast - all started in this window. First open-source model crosses 100K downloads.

  • 2024

    DPO / preference tuning at scale

    Jaskier 7B DPO family released. Production fine-tunes at SurferSEO cross 100M tokens/day.

  • 2025-2026

    Agentic systems, evals, observability

    Brand24 Slack agent ships with sub-5s p50. Internal tooling for trace replay and trajectory evals becomes a customer-facing service.

// Recognition

External signals.

  • #1 Deloitte Companies to Watch in Poland
  • 13+ first-place hackathon wins (HackYeah, Supervision Hack, BCG Platinion, others)
  • Models cited and used by HuggingFace community (~80K downloads/month, lifetime > 2.7M)
  • 10+ peer-reviewed publications across NLP, speech, and applied ML

Where

bards.ai sp. z o.o.
ul. Wielka 67
53-340 Wrocław, Poland

Registration

KRS 0000939143
NIP 8992913362
REGON 520709530

Contact

hello@bards.ai
Reply within one business day.